▲AI PRISM* Customized Economic Briefing
*Editor's Note: 'AI PRISM' (Personalized Report & Insight Summarizing Media) is an AI-based customized news recommendation and summary service developed with support from the Korea Press Foundation. It provides six curated news items tailored to each reader type.
[Key Issue Briefing]
■ Surging Token Costs: The CTO of a Korean company A alone consumed 10 million won worth of tokens in a single month, while the entire team under his supervision used up to 60 million won. As more companies run AI agents (AI systems that autonomously perform specific goals) around the clock, managing token costs is emerging as a core challenge of AI governance.
■ Spread of Token-Based Pricing: Based on OpenAI's pricing, output tokens are six times more expensive than input tokens, and major AI companies such as Anthropic and OpenAI are rapidly shifting from flat-rate pricing to usage-based token billing. Nvidia CEO Jensen Huang emphasized the importance of token consumption, saying, "If you're a $500,000-a-year engineer and you're not spending more than $250,000 on tokens by year-end, you should be worried."
■ Gap in AX Talent Policy: In February, Korea's Ministry of Science and ICT urgently convened more than 70 experts from industry, academia, and research institutes to discuss the direction of talent development for the era of agentic AI (AI equipped with autonomous execution capabilities). However, the rigid university enrollment quota system and budget-dependent education planning were cited as obstacles. KAIST's College of AI launched in March this year, but only 12 undergraduate students have applied against a quota of 100.
[News of Interest to Startup Founders]
1. "One Team Spent 60 Million Won a Month"…Korean Companies Struggle with Token Cost Burden
- Key Summary: As more companies operate AI agents 24/7, the cost of tokens (the smallest unit of data through which LLMs process information) has emerged as a key variable in corporate AI governance. LG CNS has developed and is operating a token usage dashboard that receives usage data from OpenAI with a two-day lag and reprocesses it according to its own criteria. Data tech company BigValue has separately appointed an AI administrative agent that tallies the previous day's token usage and prepares cost reports each day. AI startup Liner is preventing service development delays by issuing Lite API keys that allow developers to utilize tokens regardless of the type of LLM (large language model).
2. Output Six Times Costlier Than Input…Silicon Valley Pursues Monetization Through Token Billing
- Key Summary: Major AI companies such as OpenAI and Anthropic are shifting their revenue models from flat-rate pricing to token-based billing proportional to actual usage. Currently, based on OpenAI's pricing, input tokens cost $2.5 per million, while output tokens cost $15 per million — six times more expensive. Anthropic has converted its enterprise product "Claude Enterprise" from flat-rate to token-based pricing, charging additional fees for extra usage, and has excluded Claude Code (an AI coding tool) from its lower-priced flat plans. Meta runs an internal token usage leaderboard called "Claudionomics," periodically disclosing the top 250 users to spur internal competition and awarding titles such as "Token Legend" to top performers. Meanwhile, U.S. corporate expense management platform Ramp analyzed that "as companies shift to volatile token-based pricing plans, blind spots are emerging where finance teams can no longer track spending."
3. Uniform Standards Risk Missing Golden Time…AX Talent Policy Needs Complete Overhaul
- Key Summary: When the Ministry of Science and ICT urgently convened more than 70 experts from industry, academia, and research institutes in February, cultivating interdisciplinary talent combining domain-specific knowledge with AI utilization skills was identified as the top priority for preparing for AX (AI transformation). KAIST established its College of AI in March, but only 12 undergraduates applied against a quota of 100, and similar missteps are feared at UNIST, DGIST, and GIST, which plan to launch AI colleges next year. In the United States, MIT's Schwarzman College of Computing, launched in 2018 with a $1 billion (approximately 1.47 trillion won) investment, serves as a hub for AI education in collaboration with all undergraduate departments, while China's Yao Class (Computer Science Experimental Class) at Tsinghua University has produced numerous founders of AI unicorns, including Megvii and Xiaomazhixing. In addition, China completed the world's first national-level AI textbook development in 2018 and announced the "University AI Innovation Action Plan," building a more systematic talent development system ahead of Korea.
4. AirPods with "Eyes" Coming…AI Wearable Competition Reignites
- Key Summary: Apple is set to release AirPods with built-in cameras as early as September this year, with the cameras linked to voice assistant Siri to collect visual information and assist user decisions. Google launched "Fitbit Ace," an ultra-small, ultra-light smart bracelet without a screen, on the 7th, weighing 5.2 grams with battery life of up to seven days and a heart rate sensor, priced at $99.99 (approximately 145,000 won). Meanwhile, OpenAI is preparing voice AI devices after acquiring "io," a startup founded by former Apple Chief Design Officer Jony Ive, for $6.5 billion (approximately 9.5 trillion won) last year, and Apple is reportedly also developing smart glasses and a camera-equipped pendant. The wearable device market, which once saw fierce competition before quieting down, appears to have entered another round of intense competition combined with AI.
▶Direct link to article: "We'll Even Install Factories and Advanced Equipment"…Big Tech Courts SK hynix
▶Direct link to article: Samsung Labor-Management Negotiations Resume…"Post-Mediation" Begins on the 11th
▶Direct link to article: March Current Account Surplus Hits Record $37.3 Billion…Travel Also Turns to Surplus for First Time in 11 Years, Boosted by BTS
